A Deep Learning Approach Based on Lstm Network for Agriculture Sensor Missing Data Prediction

Smart farming, also known as Agriculture 4.0 has gained huge significant attention in the recent years, especially with its potential to address various agricultural challenges. The application of Artificial intelligence (AI) techniques, specifically the Deep Learning (DL) models, in smart-farming has proven effective for the different task of detection and prediction, leveraging their ability to generalize from trained data sets. One of the major problems facing the agriculture sector is the missed data caused by sensor failures, which can negatively impact ML models used for predictive tasks. In this paper, we propose a DL approach based on the Long Short Term Memory (LSTM) architecture for imputing the missing data from the agricultural sensors. Various agricultural data parameters captured from different sensors were used. in this work, we applied two different data preprocessing techniques, handling real missing values to eliminate the unknown data, and the data scaling to help the LSTM model to converge faster. The performance of the proposed solution was compared with three common statistical data imputation methods. the results demonstrate that the LSTM model consistently provided accurate predictions for the selected data parameters obtained from a real-world agricultural sensor experiment. A comprehensive comparison between the proposed LSTM-based solution and the selected statistical data imputation methods reveals that the LSTM model significantly outperforms the others in accurately predicting missing values. This study highlights the potential of advanced AI techniques, such as time series DL models, to significantly improve agricultural productivity and sustainability, making them crucial tools for the future of farming.